Below is a clean, actionable cheat-sheet that distills the most effective prompt techniques for Seedream 4 and Gemini 2.5 Flash-Image / Imagen — without the fluff.
Seedream 4 — Best Prompt Techniques Core principle:
Clear natural language → subject + action + environment + styleGeneral Rules Goal Technique Describe scenes naturally Sentence-based prompts, not keyword lists Apply style Mention precise style OR upload reference Add text in image Put text in "quotes" Control edits Name object, action, and parts to keep Ensure consistency Specify fixed elements when editing When image is complex Use arrows / boxes / doodles to mark targets Precision > verbosity Short and accurate beats long and ornate
Text-to-Image Templates Scenario How to prompt Real scene "A girl in a red coat walking… sunlight from left…" Logo "Design a logo for a gaming brand named 'PITBULL'…" Educational / UI / diagram Use technical wording + layout description
Editing Rules Task Best phrasing Add "Add silver earrings and necklace to girl. Match lighting." Remove "Remove the hat. Keep hair natural." Replace "Replace the largest bread man with a croissant man, same pose" Modify "Turn robots into red/yellow/green crystal, actions unchanged" Unclear area? Use bounding box / doodle
Reference Techniques Goal Key prompt structure Keep identity/style "Preserve character's hair, face, outfit design…" Transform to style "Convert to felted wool toy version, same pose" UI / layout from sketch "Match layout exactly. Add modern UI style…"
Multi-Image Rules Function Prompt style Composite "Replace subject in image 1 with subject in image 2" Outfit swap "Dress character from img1 with outfit in img2" Series "Create 6 cohesive emoji in same character style"
Gemini 2.5 Flash-Image — Best Prompt Techniques Core principle:
Write a cinematic or photographic scene , not keywordsKey Strategies Goal Best technique Photorealism Camera angle, lens, lighting, mood Stickers / icons "kawaii style, bold outline, transparent background" Logos/text "Text should say 'X' in clean sans serif" Product images Studio setup, lighting design, 45-degree angle Minimalist assets Specify negative space Comics / panels Describe scene + style + caption box text
Best Practices Technique Example Be hyper-specific "Elven armor, silver leaf, falcon-wing pauldrons" Add intent "For premium skincare brand packaging" Iterate conversationally "Make lighting warmer, keep pose" Break into steps "Step 1 background… Step 2 character…" Positive constraints "Empty quiet street" vs "No cars" Control camera "85mm lens, soft bokeh, warm sunlight" Rectify text first Generate text → then generate final image
Editing Workflows Task Pattern Add "Add small knitted wizard hat, match fur texture" Replace "Change blue sofa to brown leather, keep lighting" Style transfer "Render in Van Gogh brushstroke texture…" Multi-image "Put dress from img1 on model in img2…" Preserve details "Keep face unchanged. Place logo naturally on fabric"
Imagen vs Gemini — When to Use Which Need Use Highest photoreal / style fidelity Imagen (4 / Ultra) Best text accuracy in images Imagen (4 / Ultra) Best iterative editing / multi-turn Gemini 2.5 Flash-Image Fast creative concepts / drafts Gemini 2.5 Flash-Image
